以TiMoN冷卻工程技術提升Fe-HfO2記憶體內運算操作效率與準確性;A TiMoN-Based Cooling Solution to Enhance Efficiency and Accuracy in Fe-HfO₂ Computing In-Memory.

NCU Institutional Repository > 資訊電機學院 > 電機工程研究所 > 博碩士論文 > Item 987654321/99414

請使用永久網址來引用或連結此文件: https://ir.lib.ncu.edu.tw/handle/987654321/99414

題名:	以TiMoN冷卻工程技術提升Fe-HfO2記憶體內運算操作效率與準確性;A TiMoN-Based Cooling Solution to Enhance Efficiency and Accuracy in Fe-HfO₂ Computing In-Memory.
作者:	熊妙華;Hsiung, Miau-Hua
貢獻者:	電機工程學系
關鍵詞:	鐵電電晶體;冷卻工程;非揮發性記憶體;FeFET;CiM;NVM
日期:	2025-12-09
上傳時間:	2026-03-06 18:56:40 (UTC+8)
出版者:	國立中央大學
摘要:	為解決高頻寬記憶體（HBM）及記憶體內運算（Compute-in-Memory, CiM）系統中因熱堆積造成的焦電與極化不穩定問題，本研究提出以高熱導TiMoN電極取代傳統TiN，用以改善Fe-HfO2的熱穩定性與可靠性。實驗結果顯示，TiMoN 結構於載台110 °C下的表面溫度較TiN低約 8 °C，證實其具優異導熱能力。第一原理模擬亦指出，HfO₂/TiMoN介面溫度較TiN低50 K，可有效抑制焦電效應與熱激發載子生成，避免產生熱誘導電場。 XPS與Bader分析顯示，TiMoN電極可顯著降低介面氧空缺與次氧化物形成比例（由65%降至25%），提升HZO晶格穩定性。高溫操作下（85 °C），TiMoN-FeFET的記憶視窗僅縮減13%，遠優於TiN的34%，並能維持可逆極化行為與穩定的Pca2₁鐵電相結構。頻率與溫度掃描結果亦顯示，TiMoN結構具較低的能量損耗（DF）與介電退化率，能支撐高頻脈衝操作而不產生顯著損耗。此外，在NeuroSim平台上以MNIST資料集進行類神經運算模擬，TiMoN-FeFET在85 °C仍維持91.1%的辨識準確率，與室溫結果幾乎相同，而TiN則因導電非線性與熱漂移導致訓練效率明顯下降。綜合而言，TiMoN電極不僅改善 FeFET的熱導與介面穩定性，亦確保其在高溫下維持穩定的極化切換與權重更新，展現其於高溫類神經與CiM應用中的高度潛力。 ;To address the pyroelectric and polarization instability issues induced by thermal accumulation in high-bandwidth memory (HBM) and Compute-in-Memory (CiM) systems, this study proposes replacing the conventional TiN electrode with a high-thermal-conductivity TiMoN electrode to enhance the thermal stability and reliability of Fe-HfO₂ devices. Experimental results show that the TiMoN structure exhibits a surface temperature approximately 8 °C lower than TiN at a stage temperature of 110 °C, confirming its superior heat-dissipation capability. First-principles simulations further reveal that the HfO₂/TiMoN interface temperature is about 50 K lower than that of HfO₂/TiN, effectively suppressing pyroelectric effects and thermally activated carrier generation, thereby preventing the formation of thermally induced internal fields. XPS and Bader charge analyses indicate that the TiMoN electrode significantly reduces the formation ratio of interface oxygen vacancies and sub-oxides (from 65% to 25%), thereby improving the structural stability of the HZO lattice. Under high-temperature operation (85 °C), the memory window (MW) of TiMoN-FeFETs shrinks by only 13%, much smaller than the 34% reduction observed in TiN devices, while maintaining reversible polarization behavior and a stable Pca2₁ ferroelectric phase. Frequency- and temperature-dependent measurements also show that the TiMoN structure possesses lower energy dissipation (DF) and dielectric degradation, enabling high-frequency pulse operation without significant loss. Furthermore, neuromorphic simulations performed on the NeuroSim platform using the MNIST dataset demonstrate that TiMoN-FeFET retains a 91.1% recognition accuracy at 85 °C, nearly identical to that at room temperature, whereas TiN devices exhibit degraded training efficiency due to nonlinear conductance and thermal drift. Overall, the TiMoN electrode not only enhances thermal conductivity and interfacial stability, but also ensures stable polarization switching and weight updating under high-temperature conditions, demonstrating its strong potential for high-temperature neuromorphic and CiM applications.
顯示於類別:	[電機工程研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	20	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....